The Ryzen 7 5800X has a way to high powertarget (great results with powerlimit tweaks)

[Deleted member 193596]

i am not a native english speaker.


i started investigating many complaints about the 5800x running very hot.

in the last 14 days i tried over 200 different PBO tweaks to get the most out of my Chip.
the 5800X sits on a B550 STRIX-F and is cooled by a Kraken X73.

even with 100% Pump Speed i get like 78-85°C in Cinebench R20... and the reason is pretty clear.
the 5800X has the same power/current limits as a 5950X. (142W PPT, 140A EDC, 95A TDC)
that means the cores consume around twice the power on the 5800X.


after AMD said "this is normal and expected" i started living with it and tweaked it with PBO.

after a specific "benchmark" from me in Rainbow six siege (a custom map with the indentical paths, destroying specific walls with the same operator)

i ran every benchmark 15 times with stock, and the two most successful PBO tweaks.

Results for Rainbow Six Siege:

Stock Mainboard Settings with DOCP on.
all Cores were between 4.35 and 4.5 Ghz with an average all core clockspeed of around 4.43 Ghz. (average VCore 1.403 and 63°C) Average FPS 309

PBO PPT 180 / TDC 140 / EDC 180. 2X Scalar, 100 Mhz Offset.
all cores were between 4.45 and 4.55 Ghz with an average all core clockspeed of around 4.48 Ghz. (average VCore 1.437V and 67°C) Average FPS 313

PBO PPT 200 / TDC 165 / EDC 200. 4X Scalar, 150 Mhz Offset.
all cores were between 4.45 Ghz and 4.58 Ghz with an average all core clockspeed of around 4.52 Ghz. (average VCore 1.465V and 72°C) Average FPS 314


Results with Cinebench R23:
Stock:
1609 SC / 15533 MC (78.8°C 1.331V @ 139W)
4850 Mhz / 4500 Mhz

PBO PPT 180 / TDC 140 / EDC 180. 2X Scalar, 100 Mhz Offset:
1623 SC / 15549 MC (81.5°C 1.34V @ 144W)
4950 Mhz / 4525 Mhz

PBO PPT 200 / TDC 165 / EDC 200. 4X Scalar, 150 Mhz Offset:
1641 SC / 15602 MC (86.1°C 1.365V @ 153W)
4975-5000 Mhz / 4550 Mhz


then i started taking some Power off the CPU and let it do it's thing. (Using PBO with the following settings reduced the performance as much as -200 MHz)

DOCP + Changing Powerlimits in PBO without changing any Scalar or Clockspeed Offset.

PBO PPT 105 / TDC 70 / EDC 90

Rainbow Six Siege:

all cores were between 4.58 Ghz and 4.72 Ghz with an average all core clockspeed of around 4.67 Ghz. (average VCore 1.38V and 61°C) Average FPS 324

Cinebench R20:
1611 SC / 15091 MC (64.8°C 1.2V @ 105W)
4850 Mhz / 4425 Mhz



The results are the biggest improvement so far.
Gaming and other lighter Tasks are much faster than PBO while consuming way less power.
Only the Multicore performance is around 3.5% lower while consuming around 33% less power.

i am currently trying this in other applications like Blender and other/heavier Videogames like Battlefield V and Cyberpunk 2077.

i highly recommend testing this for yourself

 

[dont whant to set it"']

Nice and thanks for sharing.

On mine , I found the best way to lower power consumption was a manual setting for the vcore . Last time it actually posted and booted was set to 1.23V@ 4.525MHz all-core oc , with wich settings I gamed and ran a couple of tests. Its sported by standard issued AM2 cooler mind you.
It no Post and boot no more, but hopefully, I get a new and different mb delivered tomorrow for a full refund on my Strix B550-F.

 

[Deleted member 193596]

Here are my Clock Speeds with the power tweak in CoD Warzone (CPU sits at around 70% Load)
that's around 200-250 Mhz more than Stock.

 

[bluefrisky]

Hey, any update on this?

I followed your steps and ended up reducing PPT to 125, EDC to 125 and TDC to 85. I also used the curve optimizer with a negative offset of 10 on all cores, and increased frequency boost in PBO by 100mhz.

I’m now getting about the same multi score in R23/CPU-Z as my stock scores, and higher single core performance, and I’m running about 8C cooler (from 87 peak in R23 to 79).

When I drop the numbers as drastically as you did, my performance tanks, but it still helped me out a lot! No longer thermal throttling and temps never go above 80 in any scenario, better single core performance than stock, and almost the same multi score! Thank you!

 

[Deleted member 193596]

Hey, any update on this?

I followed your steps and ended up reducing PPT to 125, EDC to 125 and TDC to 85. I also used the curve optimizer with a negative offset of 10 on all cores, and increased frequency boost in PBO by 100mhz.

I’m now getting about the same multi score in R23/CPU-Z as my stock scores, and higher single core performance, and I’m running about 8C cooler (from 87 peak in R23 to 79).

When I drop the numbers as drastically as you did, my performance tanks, but it still helped me out a lot! No longer thermal throttling and temps never go above 80 in any scenario, better single core performance than stock, and almost the same multi score! Thank you!

when i use this "tweak" with PBO enabled it won't work properly.

i ended up using this and -5 on the curve optimizer.

 

[INSTG8R]

when i use this "tweak" with PBO enabled it won't work properly.

i ended up using this and -5 on the curve optimizer.

The right way do It. Hoping GB releases the BIOS with it before my 5600X gets here next week

when i use this "tweak" with PBO enabled it won't work properly.

i ended up using this and -5 on the curve optimizer.

what kind of max voltage did that get you I’m curious how “low” you can go I’ve see 12 or 15 being used but just wondering how it translates as an undervolt I currently run my 3700X with a -.1500 offset which gets me 1.35 and I know I could go even lower but I’m under water so no need to test the limit. This Curve Optimzer will make it even easier but how much -5 for example takes off your previous 1.45V

 

[brandofriva]

Thanks very much for this info in this thread. At stock my CPU was running at 90degC even on custom water cooling. Managed to cut a few degrees from my temps with this while not sacrificing too much performance using some of the settings above.
Are there any PBO guides people here can recommend to finetune this for my CPU/setup properly? (either to reduce temps and/or improve performance)

 

[Deleted member 193596]

The right way do It. Hoping GB releases the BIOS with it before my 5600X gets here next week


what kind of max voltage did that get you I’m curious how “low” you can go I’ve see 12 or 15 being used but just wondering how it translates as an undervolt I currently run my 3700X with a -.1500 offset which gets me 1.35 and I know I could go even lower but I’m under water so no need to test the limit. This Curve Optimzer will make it even easier but how much -5 for example takes off your previous 1.45V

i checked my VCore with three different settings. and the new 1.1.9.0 AGESA

Stock
"Powertweak with 105W"
and with Curve Optimizer.

Black Ops Cold War solo custom match at 480P with a 3090 forces my CPU to it's limits and the clock speed is very stable.

Stock:
~4.55 Ghz all core with one core at 4.65 Ghz at 1.412V

Powertweak:
~4.65 Ghz on all Cores and two cores at 4.7 GHz at 1.38V

Stock with Curve Optimizer at -10:
~4.7 Ghz on all Cores (4680-4700 Mhz) at 1.394V

Powertweak plus Curve Optimizer at -10:
far above 4.7 Ghz on all cores except one and ~4.8 Ghz on two cores. at 1.378V


The Curve Optimizer is stable up to -10 across all cores.
at -15 i start getting crashes in games like Battlefield or Rainbow Six Siege (no bluescreens but it is absolutely on the edge)

 

[bluefrisky]

when i use this "tweak" with PBO enabled it won't work properly.

i ended up using this and -5 on the curve optimizer.

I decided to follow in your footsteps more closely.

I lowered PPT to 115, EDC to 100 and TDP to 80, then applied -5 to all cores in curve optimizer. I am seeing incredible results.

My multi core is down 3%, single core is the same, tested in both CPU-Z and R23 multiple times. My peak temperatures during full stress load dropped by about 13c, I am now never going above 70c in R23.

I tested stability for hours across many bench tests and games, it seems perfectly stable.

As for gaming performance, so far CP2077 and Shadow of The Tomb Raider are showing slightly increased FPS (including better 1% lows).

I think you are right, the 5800x draws way too much power and overheats itself for no apparent reason.

I also discovered something interesting recently. The 5800x is unique in the way it generates heat, and may require adjustment of the positioning of the cooler and mounting. Read more here:

https://www.reddit.com/r/overclocking/comments/kszoh9

 

[Zach_01]

Comparing a 5800X to 5900X/5950X the main difference is that while all 3 have the same power consumption and heat dissipation, on 5800X all heat comes from a single CCD. Thus it has double heat density. That’s not an easy task to accomplish when trying to maintain temperature to a reasonable limit. Der8aur’s solution may help a little but don’t expect miracles.
It also involves cooler dissipation capacity and thermal paste/Tim material for transferring heat quickly.

 

[bluefrisky]

Comparing a 5800X to 5900X/5950X the main difference is that while all 3 have the same power consumption and heat dissipation, on 5800X all heat comes from a single CCD. Thus it has double heat density. That’s not an easy task to accomplish when trying to maintain temperature to a reasonable limit. Der8aur’s solution may help a little but don’t expect miracles.
It also involves cooler dissipation capacity and thermal paste/Tim material for transferring heat quickly.

Interesting. I have read some reviews and most claim 5-7c lower temps, I’d say that’s pretty good.

 

[Zach_01]

Like I said, it involves cooler capacity and TIM heat transfer rate. 6-7C reduction might be the best case scenario with open loop waterblock or highend AIO with special TIM, other than regular paste.

 

[Leothelesser]

Curve optimizer with a negative offset of -30 on all cores
PPT to 120 EDC auto and TDC auto
Chinebench20 6069 multi 625 single 78c

Offset -30 PPT auto
Chinebench20 6145 multi 624 single 86c

Offset -30 PPT 135
Chinebench20 6117 multi 624 single 85c

Offset -30 PPT 125
Chinebench20 5957 multi 625 single 80c

Offset -30 PPT 120
Chinebench20 6069 multi 625 single 78c

Why did the PPT limit dropping from 125 to 120 gave an increase to multi core score?

 

[Zach_01]

Offset -30 PPT 125
Chinebench20 5957 multi 625 single 80c

Offset -30 PPT 120
Chinebench20 6069 multi 625 single 78c

Why did the PPT limit dropping from 125 to 120 gave an increase to multi core score?

Do you remember speed, EDC and voltage values during those test 2 tests?

PPT alone can’t say much.

 

[bluefrisky]

Curve optimizer with a negative offset of -30 on all cores
PPT to 120 EDC auto and TDC auto
Chinebench20 6069 multi 625 single 78c

Offset -30 PPT auto
Chinebench20 6145 multi 624 single 86c

Offset -30 PPT 135
Chinebench20 6117 multi 624 single 85c

Offset -30 PPT 125
Chinebench20 5957 multi 625 single 80c

Offset -30 PPT 120
Chinebench20 6069 multi 625 single 78c

Why did the PPT limit dropping from 125 to 120 gave an increase to multi core score?

A lot of this might be run to run variance. Sometimes with the same values I get a score that’s higher or lower by as much as 100 points. You need to do multiple tests and average them out.

Anyway, your temps and score look good with the -30 curve and 120ppt, I’d stay on that. Should be nice in games too.

 

[Leothelesser]

TDC 82.608a
EDC 135.507a
Core Voltage 1.325v
Core VID Effective 1.350v

4830.2 mhz
4552.2 average effective

 

[Zach_01]

TDC 82.608a
EDC 135.507a
Core Voltage 1.325v
Core VID Effective 1.350v

4830.2 mhz
4552.2 average effective

Over which of the 2 runs in question?

You need to compare these values over both runs (120 and 125 PPT) to come to a conclusion about why lower PPT gave better score.

 

[blu3dragon]

I'm skeptical. Why would lowering the power target make the cpu faster?

Cooler, I get, but faster doesn't seem right. There's either something else going on or AMD got something wrong here.

 

[bluefrisky]

I'm skeptical. Why would lowering the power target make the cpu faster?

Cooler, I get, but faster doesn't seem right. There's either something else going on or AMD got something wrong here.

It makes perfect sense. Without a lower power target, the CPU’s hit 80c or higher (in a lot of cases 90c) and start down clocking. If you’re running at 75c or lower, you’ll have higher sustained clocks.

 

[ratirt]

I'm thinking about the HWUB (GN as well BTW) Youtube channel when they have specifically said, lowering power targets to make a CPU clock run higher, doesn't always translate to a better performance. (Obviously you will have lower temps as well) I only hope you keep that aspect in mind while you fiddle with the settings.

 

[blu3dragon]

It makes perfect sense. Without a lower power target, the CPU’s hit 80c or higher (in a lot of cases 90c) and start down clocking. If you’re running at 75c or lower, you’ll have higher sustained clocks.

I dunno. The 5800x has a stock thermal limit of 90c. It appears to adjust clock and voltage to stay below that. However, if it hits a power limit I would expect it to do exactly the same thing. It will need to down clock and down volt to stay under the limit.

 

[mtcn77]

It makes perfect sense. Without a lower power target, the CPU’s hit 80c or higher (in a lot of cases 90c) and start down clocking. If you’re running at 75c or lower, you’ll have higher sustained clocks.

It is because of PTTL, right in the same drop menu that you find the 'scalar' setting.
The question is to set as low as possible such that even if the cpu was prime 95 tested it wouldn't trigger the temperature threshold. The issue is finding a setting which won't trigger it and then setting up a threshold that it will boost over it, but then as maximum temperature is reached it will coast alongside it. In summary, you keep PPT tidy along with EDC at the beginning and then delimit PPT. I suggest decreasing EDC from its 140 default to 130 such that there is a 10% gap from PPT which was a key point in former Ryzen generations.
An easy way to find out is prime95. As it scales in power output according to the sram usage, if you use less FFT to do the calculations, you will be effectively singling out the execution unit stability.
I found good quotes on that over the net;

For example, by selecting 8-8 kB as the FFT size, the program stresses primarily the CPU.

Thus a 4K FFT will use 32KB of data plus some for sin/cos and other data. It would probably fit in 64KB of L2 cache.

 

[Leothelesser]

Offset -30 PPT 120
Chinebench20 6069 multi 625 single 78c

My idea was to reduce my temps as my cooler is only a be Quiet! Dark Rock Slim and the ambient temp is 30c + all summer.

The post talked of a “way to high powertarget” so I limited the power target and set a -30 offset to see if I could post. Then I started lowering the PPT with everything else on Auto and when I got the above 6069 and 625 I was happy with that.

Not the best stats but Reasonable and with Better temps.

 

[Nexium]

Hello Everyone,

I was looking around the web for some current 5800x PBO2 Curve Optimizations out there to compare to my own results and experience. Not too much out there yet as things are still fairly new. I'd thought I'd contribute here with my results while I'm at it. Keep in mind I'm no expert, just a person that also likes to find the CPU limits and squeeze max CPU performance while maintaining stability and comfortable system temperatures for gaming. I mainly use Cinebench and various games to test temperatures and stability. If someone has suggestions what else I can try to tweak it further for higher stable boost clocks through the MSI BIOS settings, I'm all ears!

From what I understand, these Ryzen 5000 series PBO2 method of overclocking is to find the CPU "Zen" to allow the CPU's hardware coded power algorithm to hit the highest boost speeds often and consistently is the goal which is different than the conventional way of all core overclocking that I was used to with past CPUs over the years. In my observations using HWinfo64 monitoring, while my 5800x can boost 5.1Ghz on some cores with -20 All Core Curve, it doesn't boost to that speed very often or consistently and of course unstable. As I experimented moving down the boost clock speeds by -25Mhz from 5.1Ghz until 5Ghz that's when I noticed my average clock speeds increased as the CPU can boost to top speeds more often. This leads me to believe if the CPU struggles at certain overclocked boost speeds it reluctantly or will seldom hit your top boost clock speeds based on its algorithm to ensure system stability as top priority.

I found my 5800x is most happy and stable so far in both Cinebench R23 with various games all cores boosting 5005Mhz consistently with the BIOS settings I provided below on a MSI Unify x570 with the beta BIOS 7C35vA86 (1/12/2021).

One other thing to note here is I was only initially stable at -12 Curve All Core Offset at 5Ghz and any offset values of more than -12 would crash to desktop in any games. Adding CPU voltage offset and LLC seemed to allow me to stabilize at my current -15 Curve All Core Offset and at the same time allowed the CPU to hit my top boost clock speeds even more consistently and often. Again, my theory is this should have allowed the CPU to work with the lower power curve allowing more frequent max boost clocks and the motherboard provides the CPU voltage offset in order to stabilize voltages at the CPU lower power curve working in tandem.

PBO = Advanced
PBO Limits = Disabled
PBO Scalar = x7
Max CPU Boost Clock Override = +150Mhz
Platform Thermal Throttle Limit = 255
Curve Optimizer = All Cores -15
Bus Clock = 100.1Mhz
CPU Offset Voltage = 0.0125v
CPU LLC = Level 3

System Temp Average: 27C
All Core Max Temp: 76C
Single Core Max Temp: 62C

Cinebench R23 Scores
MultiCore: 15601
SingleCore: 1653

System Specs:
Seasonic Prime 750w
MSI Unify x570 Beta BIOS 7C35vA86 (1/12/2021)
Ryzen 5800x
Phanteks 280mm AIO
G.Skill Neo 32GB Dual Channel 16-19-19-39 1T @ 1.35v- XMP
EVGA 1080Ti FTW3
Sabrent Rocket Plus 2TB NVME SSD
Samsung 970 Evo 1TB NVME SSD

 

[mtcn77]

PBO Scalar = x7

CPU LLC = Level 3

These two do the same thing. You can select one or the other. Ryzen doesn't need you to set voltages, it can demand it from the motherboard itself. Try to keep one and forego the other. It misses the point if you reduce 15mV and add 125 somewhere else. It only reduces base voltage and that is without the incrementally static 125 offset anyway.
We dearly miss the Stilt making things clearcut.